Artificial intelligence-guided design of LNPs for in vivo targeted mRNA delivery via analysis of the spatial conformation of ionizable lipids.

Efficient mRNA delivery to specific tissues requires optimized ionizable lipids, yet the role of lipid spatial conformation in organ targeting and endosomal escape remains underexplored. Here we developed a library of lipids with diverse amino heads, degradable linkers and hydrophobic tails, generating distinct three-dimensional conformations. Molecular dynamics simulations revealed the dynamic conformations of these lipids during organic-aqueous phase transitions, and experimental validation confirmed that head and tail arrangements are key determinants of delivery efficiency and organ specificity. To accelerate lipid discovery, dynamic conformation data were converted into 2D density images to train machine learning models for lipid selection. AI-guided candidates, notably lipid P1, adopted stable three-tail cone-shaped conformations that promoted IgM protein corona formation and enabled spleen-targeted mRNA delivery. In preclinical models, P1-based mRNA vaccines triggered strong antibody and T-cell responses, leading to marked tumour suppression. These results highlight the pivotal role of lipid spatial conformation and the potential of AI-driven strategies to optimize lipid nanoparticles for organ-specific mRNA delivery.